Illuminating Neural Circuits in Alzheimer’s Disease

Abstract

Alzheimer’s disease (AD) is the most common neurodegenerative disorder and there is currently no cure. Neural circuit dysfunction is the fundamental mechanism underlying the learning and memory deficits in patients with AD. Therefore, it is important to understand the structural features and mechanisms underlying the deregulated circuits during AD progression, by which new tools for intervention can be developed. Here, we briefly summarize the most recently established cutting-edge experimental approaches and key techniques that enable neural circuit tracing and manipulation of their activity. We also discuss the advantages and limitations of these approaches. Finally, we review the applications of these techniques in the discovery of circuit mechanisms underlying β-amyloid and tau pathologies during AD progression, and as well as the strategies for targeted AD treatments.

Introduction

Alzheimer’s disease (AD) is the most common neurodegenerative disorder in the elderly. The characteristic clinical symptom is progressive memory loss [1]. Several key neuropathological hallmarks of AD have been revealed, including extracellular neuritic plaques composed of β-amyloid (Aβ) and intracellular neurofibrillary tangles (NFTs) composed of hyperphosphorylated tau protein [2]. Both Aβ plaques and NFTs impair synaptic function and neural circuit networks. Using functional MRI, electrophysiological techniques, and biochemical data, researchers have found aberrant neural activities in patients with AD and mouse models [3, 4]. These abnormal activities trigger a disorganized brain network and interfere with the intricate processes underlying learning, memory, and other cognitive functions. Pharmacological therapies targeting either Aβ or tau have shown limited efficacy in the treatment of AD. This has motivated novel approaches designed not only to specifically rescue the cognitive impairments in AD but also avoid drug-induced side-effects or pharmacoresistant symptoms.

Different types of brain cells orchestrate to modulate cognitive function. Heterogeneous brain cells show distinct alterations during AD development. Given the advantages of neuron type-specificity and manipulation accuracy, single-cell RNA sequencing, optogenetics, and chemogenetics are increasingly used in the study of AD to investigate molecular and circuit mechanisms. Novel findings from AD studies that use these cutting-edge techniques may help us better understand the AD memory dysfunction induced by Aβ and tau proteins. They may also provide novel therapeutic strategies for AD treatment in the future. In this review, we briefly describe the most advanced tools currently available for neural circuit research relating to AD.

Single-Cell RNA Sequencing

There are billions of highly differentiated and interconnected cells within the brain. Distinct neurons communicate with each other and form intricate neural networks to control physical and mental functions. Traditional research at the bulk-tissue level of resolution may mask the complexity of changes across brain cells and within cell groups [5]. For example, using bulk RNA-sequencing, it may be difficult to differentiate whether low-abundance transcripts are expressed at low levels in common cell types or at high levels in rare cell types. Moreover, in diseases, the low abundance transcriptional changes in a rare cell type may be undetectable when a bulk RNA-sequencing approach is employed. As such, the newly developed single-cell RNA sequencing (scRNA-seq) techniques have the potential to overcome these limitations. In addition to providing a comprehensive landscape of brain cell-type diversity based on transcriptional profiles, the scRNA-seq method may be useful in mapping the mammalian connectome and exploring the mechanisms of brain-related disorders, such as AD, at the single-cell level.

Individual cells can be physically separated. The isolation methods vary in the number of isolated cells (high-throughput or low-throughput) and the means of cell selection (biased or unbiased) [6]. By whole-cell patch clamping, cellular content from a sucked single cell can be used to acquire RNA for quantitative RT-PCR or microarray analysis [7, 8]. Moreover, this method provides additional information, such as location, electrophysiological properties, and morphology of the examined cell. However, it is not quantitative enough, nor does it provide an unbiased assessment of global gene expression patterns. Moving from pipetting and manual selection, several automated single-cell compartmentalization methods have been developed, such as fluorescence-activated cell sorting and the fluidigm C1 system. Although they have been combined with scRNA-seq techniques, including SMART-seq and CEL-seq, to explore the brain at the single-cell level [9–11], there are two limitations: (1) the total number of captured cells is not quantitative enough for the high-throughput format; and (2) they do not maintain the native heterogeneity of brain constituents [12, 13]. Currently, droplet-based technologies to randomly capture single cells with barcoded beads are commonly used as a high-throughput and unbiased solution [14]. The three most widely used platforms are Droplet sequencing (Drop-seq) [15], indexing droplets RNA sequencing (inDrop) [16], and GemCode/Chromium 10× (widely known as 10× Genomics) [17, 18]. They use microfluidics to tag individual cells with single beads containing a unique barcode. Each mRNA transcript is also linked with a unique molecular identifier (UMI). By comparison, 10× Genomics is the most sensitive for detecting the greatest number of transcripts, while inDrop is ideal for detecting weakly-expressed genes (DEGs) [17].

scRNA-seq techniques require freshly-harvested, viable, and intact cells. Moreover, most of the single-cell isolation methods may impair neurons and cause some cell types to be under- or over-represented in the final datasets. Since nuclei are more resistant to mechanical and physical stress, single nuclei RNA sequencing (snRNA-seq) is an alternative approach to scRNA-seq and has been used in the study of the cellular composition in the brain [19–22]. Compared with whole-cell transcriptomes, there is no difference in snRNA-seq data in terms of the number of detected genes and resolved cell types [23].

Full-length and tag-based approaches are used to generate single-cell sequencing libraries [24]. Since reads are derived from across the entire length of genes, full-length methods greatly improve the overall sensitivity. It is worth noting that full-length library preparation shows a bias for longer genes rather than shorter genes because counts are often missed for the latter [25]. In tag-based methods, the incorporated UMIs allow the identification and quantification of individual transcripts.

A generalized single-cell transcriptomic experiment includes single cell isolation, barcoding, cDNA amplification, library construction, sequencing, and analysis. For each experiment, the technical performance of the scRNA-seq data should be evaluated from the sensitivity, accuracy and precision aspects [24, 26, 27]. Sensitivity measurement is highly dependent on the depth of sequencing [24, 26]. Accuracy can be influenced by factors specific to the protocol being used [24, 26]. Precision is inversely proportional to the technical noise in the RNA-seq measurements [24, 26].

A major disadvantage of the scRNA-seq method is the lack of spatial information [28–30]. By fluorescence in situ hybridization (FISH), the spatial locations of genes identified in scRNA-seq can be clearly visualized. For example, in single-molecule FISH (smFISH), multiple fluorescent probes are used to characterize distinct cell groups [31]. However, the number of colors that can be visualized at once is very limited in this technique. In contrast, sequential FISH (seqFISH) can sequentially label tissue for different RNA markers [32, 33]. After each round of labeling, fluorescence is washed out and cell populations are binarily labeled for the presence or absence of a particular gene. As such, seqFISH is more efficient and robust in cell population identification than smFISH.

Cutting-Edge Experimental Approaches for Visualizing Neural Circuits

To understand the complex brain functions carried out by neural circuits, it is necessary to dissect their structural organization. To date, several viral tracers have become available to delineate the structural connectivity between distinct neurons or different brain regions

Anterograde Trans-Synaptic Tracing

Herpes simplex virus (HSV) and vesicular stomatitis virus (VSV) have been commonly used in anterograde trans-synaptic tracing. Both viruses invade and replicate in first-order infected starter neurons and then spread into synaptically connected neurons.

HSV

The HSV type 1 strain H129 has been widely used in anterograde multi-synaptic tracing [34–36]. Tyrosine kinase (TK) is required for HSV replication and is necessary for its anterograde spread in non-mitotic cells. Recently, HSV has been genetically modified to achieve monosynaptic tracing. In this system, TK is deleted and then trans-complementation of TK restores the ability of H129-ΔTK to replicate and spread into directly projected targets. If the expression cassette of the TK gene is in a Cre-recombinase-dependent strategy, the starter cell type can be greatly restricted.

H129 mainly transmits anterogradely, but it can also infect upstream neurons through axon terminal uptake [37]. More importantly, the main drawback of the H129 strain is its high toxicity. It can also cause cell death and animal death. With current trans-multisynaptic H129 tracer systems, experimental animals die within 3 to 5 days after injection of H129 [38], which greatly restricts the sampling time to within 3 days. Although toxicity can be reduced by engineering viral genomes, such as TK-null H129 [39], this problem has not been resolved completely. TK-deficient recombinant viruses show milder cytotoxicity without damage to their infectivity and viral gene expression. However, trans-monosynaptic H129 tracer systems remain insufficient for electrophysiological recordings from brain slices. Thus, to support functional studies, more efforts are needed to reduce the cytotoxicity of trans-synaptic HSV in the future.

VSV

VSV is an arthropod-borne virus. It leads to vesiculation and ulceration around the mouth, teats, and hoofs after infection. These symptoms usually resolve within a few weeks without fatality [40]. Compared to HSV, VSV is less virulent to humans [41].

The glycoprotein (VSV-G) is necessary for VSV binding to target cells [42]. When VSV-G binds to phosphatidylserine on the cell surface membrane [41], VSV can enter cells and start to replicate rapidly. The first progeny of viruses are generated and released within 1.5 h [43]. It has been found that the direction of VSV transmission greatly depends on the glycoprotein. The intrinsic glycoprotein endows its characteristic of anterograde transmission [44]. However, G protein from either lymphocytic choriomeningitis virus or Rabies virus can change the transmission from anterograde to retrograde [44]. Toxicity and inefficiency are the main shortcomings of VSVs. Thus, the modification of virulence and efficiency of spread will enable the wider applications of VSV in circuit structure dissection and functional studies.

A recent study showed that adenovirus associated virus 1 (AAV1) and AAV9 can be used as monosynaptic anterograde tracers [45]. They exhibit anterograde trans-synaptic spread properties with relatively low efficiency. Therefore, the viral spread is restricted to direct postsynaptic targets of the infected cells, and the labelling of postsynaptic cells relies on reporter expression activated by a recombinase-expressing virus. Such a strategy can be used to perform afferent-dependent tagging of postsynaptic neurons and support functional manipulations with Cre-dependent tool genes. However, care should be taken when using it to map connections between reciprocally connected regions because retrograde trans-synaptic spread can also occur, especially at high titers [46].

Retrograde Trans-Synaptic Tracing

Rabies Virus (RV) is a neurotropic virus. It retrogradely transmits from the infected peripheral site to the central nervous system, leading to lethal zoonotic disease. Unlike HSV, the RV spreads exclusively in the retrograde direction. High neuronal specificity without causing glial cell infection makes RV an ideal tool for neural circuit tracing.

RV envelope protein G is necessary for neuronal infection and trans-synaptic transmission of the RV. SAD-B19, an attenuated vaccine strain of RV, was developed for the monosynaptic trans-synaptic system [47, 48]. In this system, the RVG gene is replaced by EGFP and pseudotyped with envelope glycoprotein (EnvA) from avian sarcoma-leukosis virus (ASLV-A). TVA is the cognate receptor of EnvA and does not exist in mammals. Thus, the recombinant RV [SADDG-EGFP(EnvA)] cannot infect mammalian neurons unless they are exogenously supplied with TVA. RV-DG(EnvA), a mutant rabies virus with glycoprotein G gene deletion and a helper virus (AAV) that expresses TVA and RVG in a Cre-dependent manner are both injected into the desired region, where RV can only invade cells with Cre expression and then retrogradely spread to their direct upstream neurons, but cannot spread any further. Therefore, in such cases, only monosynaptic connections to starter cells are marked [49].

Recently, Schwarz et al. developed a method termed TRIO (for tracing the relationship between input and output) and cell-type-specific TRIO (cTRIO) [50]. In this system, canine adenovirus type 2 (CAV2) is used to provide the recombinase (Cre or Flp). Although CAV2 is not a trans-synaptic tracer, it can transduce neurons through their axonal terminals and efficiently transport back to the soma. In the TRIO system, after the injection of CAV-Cre into one of the output regions, Cre is specifically expressed in neurons that project to the CAV2-injected region. Then, by injecting RVDG(EnvA) and Cre-dependent AAV helper virus that expresses TVA and RVG into the targeted area, RV can only infect neurons in the targeted area which directly project to the CAV2-injected regions. It then transports to monosynaptic upstream neurons. In contrast to TRIO, cTRIO has the advantage of being neuron type-specific. CAV2 is used to express Flp and AAV is used to express TVA. RVG is only in neurons with both Cre and Flp. With the help of these two viruses, RV can infect Cre-expressing neurons in the targeted area that innervate neurons within the CAV-injected region (Fig. 1). Thus, cTRIO enables monosynaptic tracing from a specific type of cell projecting to a specific brain area.

Fig. 1.

Schematic of cTRIO for neural circuit tracing. In X-Cre mice, CAV2-DIO-Flp is injected into region C, while AAV-fDIO-TVA/G and EnvA-RV-ΔG are injected into region B. Only the presynaptic inputs to neurons that are located in region B, project to region C, and express geneX (orange cell in region B) will be retrogradely labelled by RV (green cell in region A) (A → B^X+ → C). However, presynaptic inputs to region B neurons that innervate region C but do not express gene X (grey cell in region D) (D → B^X− → C), or region B cells that do not project to region C (B → E) will not be labeled.

Cytotoxicity is the main limitation of the above RV tracing system (first generation). Chatterjee et al. recently introduced a new class of double-deletion-mutant rabies viral vectors, which show low toxicity in transduced cells [51]. They deleted a second gene known as ‘large protein’ gene (NCBI symbol: RABVgp5). This gene encodes the viral polymerase, which is required for the transcription of viral genes and replication of the viral genome [52]. Compared to first-generation viruses, these new (‘second-generation’ or ‘Δ GL’) viral vectors can also retrogradely infect projection neurons. The target neurons remain alive and healthy in terms of both electrophysiology and morphology up to a year after the virus injection. Thus, this system is promising for long-term tracing, functional detection, and optogenetic manipulation.

Cutting-Edge Experimental Approaches for Manipulating Neural Circuits

Optogenetics

Optogenetics could be an invaluable tool in neuroscience research. It encompasses the knowledge of optics, microbial biology, virology, and biochemistry [53, 54]. About 50 years ago, light-activated proteins were first discovered. Bacteriorhodopsin was discovered in 1971 [54], halorhodopsin in 1977 [55], and channelrhodopsin in 2002 [56]. Upon light exposure, these opsins act as channels or pumps to directly induce electrochemical signaling in cells. This feature is distinct from rhodopsin, which indirectly transduces electrical current via G-proteins. In 2005, a microbial opsin gene was first introduced into neurons and it precisely controlled neuronal spiking upon light exposure [57]. In 2007, the fiber-optic interface and single-component control of freely-moving mammals were described [58, 59]. A series of Cre recombinase-dependent opsin-expressing viruses have been developed and widely used in mouse lines that selectively express Cre recombinase in defined cell types. Thus, researchers can use these latest optogenetic controls to investigate the functions of defined cells in specific brain regions under physiological conditions and their alterations in diseases.

Channelrhodopsins, derived from Chlamydomonas reinhardtii, are one kind of light-driven proteins. Channelrhodopsin 2 (ChR2) is composed of seven transmembrane helices, and a retinal component is embedded within the helices [60]. Upon illumination, the retinal component absorbs light and quickly causes a conformational change, which opens the pore and allows the movement of cations down their electrochemical gradient on ChR2-expressing neurons. These inward currents drive fast membrane depolarization and robustly excite neurons [61, 62] (Fig. 2A). H134R, a gain-of-function mutant of ChR2, causes larger stationary photocurrents than wild-type ChR2 [63]. However, the recovery from inactivation of both wild-type ChR2 and H134R is slow, which limits the precision in the optogenetic control of neurons, especially at stimulating frequencies above the γ range [57, 64]. Gunaydin et al. designed point mutations in the ChR2 sequence and developed ChETA, which accelerates channel closing and rapid repolarization of neurons following light stimulation [65]. Recently, many variants have been developed, such as red-shifted (VChR1 [66], ReaChR [67]) and blue-shifted (CheRiff) variants [68].

Fig. 2.

Schematics of neural manipulation by optogenetics and chemogenetics. A, B Upon light stimulation, Na⁺ and Ca²⁺ move down their electrochemical gradients on the ChR2-expressing neurons and depolarize the membrane, resulting in neuron excitation. While hyperpolarizing current on NpHR-, Arch- or mutated ChR2-expressing neurons leads to neuron inhibition. C, D In the presence of CNO, hM3Dq-expressing neurons are activated, while hM4Di-expressing neurons are inhibited via changing the concentration of second messengers.

In addition to depolarizing cells, some microbial opsins can hyperpolarize cells upon illumination. NpHR isolated from Natronomonas pharaonis is sensitive to yellow-green light. Light activation of NpHR produces an inward Cl^– (hyperpolarizing current), which inhibits the action potential firing of NpHR-expressing neurons (Fig. 2B). Arch is a light-activated proton pump bacteriorhodopsin [69]. By pumping H⁺ out, Arch also hyperpolarizes cells when exposed to light (Fig. 2B). Furthermore, ChR2 can also be used to inhibit neurons upon light stimulation when its permeability is changed to Cl^– instead of cations after mutation in its pore structure [70] (Fig. 2B).

Chemogenetics

Recently, chemogenetics has been widely employed in neuroscience research to regulate the activities of specific neurons or neural circuits. Chemogenetics refers to the use of genetically-engineered receptors that interact with specific synthetic ligands or molecules to alter cellular signal transduction [71]. Since the receptors (e.g. G protein-coupled receptors and ligand-gated ion channels) are modified through random or site-directed mutagenesis, they are no longer responsive to their natural ligands, but can be specifically activated by synthetic chemicals [71].

Designer receptors exclusively activated by designer drugs (DREADDs) [72] are engineered G protein-coupled receptors (GPCRs) that can precisely control GPCR signaling pathways (for example, Gq, Gs, and Gi). Currently, hM3Dq and hM4Di are the most commonly-used DREADDs [73, 74]. Due to their mutations, both hM3Dq and hM4Di can only be activated in the presence of clozapine N-oxide (CNO), a compound modified from clozapine (Fig. 2C, D). CNO only activates these “designer receptors” without changing the functions of endogenous receptors. However, this specificity can be changed at a high dose of CNO. It has been reported that intracranial injections of CNO at micromolar concentrations (10 μmol/L) competitively inhibit binding at several receptors, including muscarinic M1, M3, M4, histamine H1, 5-HT2A, and dopamine D1 and D2 [75]. Thus, these off-target effects at endogenous receptors may significantly confound the agonistic effects of CNO.

Compared to optogenetics, chemogenetics has an incomparable advantage in that the stimulus can be administered via less- or non-invasive routes, such as intraperitoneal injection and oral administration [73, 76, 77]. Although chemogenetics cannot achieve precise temporal control, they can provide prolonged manipulation of neurons and their circuit activities [77].

Neural Circuit Defect in AD Brains and Strategies for Targeted Manipulations

The hippocampus is vulnerable in patients with AD [78–80], and its dysfunction is closely associated with cognitive impairment during AD progression [80, 81]. Next, we focus on the hippocampus and briefly introduce some recent studies on hippocampus-associated neural circuits in AD.

Application of Single Cell Transcriptomics in AD

In addition to cell-type classification under physiological conditions, snRNA-seq has recently been used in AD to understand the vulnerability of different cell types [19, 82]. 80,660 droplet-based single-nucleus transcriptomes from prefrontal cortex in human subjects with varying degrees of AD pathology were profiled and analyzed [19]. Differentially-expressed genes between AD-pathology and no-pathology groups were downregulated in excitatory (Ex) and inhibitory (In) neurons, while they were upregulated in oligodendrocytes (Olis), astrocytes (Asts), and microglia (Mic), indicating a heterogeneous response to AD pathologies among cell types [19]. In cellular subpopulations, many subclusters of Ex and In neurons, Asts, Mic, Olis, and oligodendrocyte progenitor cells (Opcs), have been identified. Among them, some cellular subpopulations were associated with cells isolated from subjects with AD pathology traits, while some were associated with cells from subjects with no pathology [19]. Furthermore, genes associated with protein folding and stability, neuronal and necrotic death, T-cell activation, and immunity were identified in some subpopulations of AD pathology-associated In neurons, Opcs, and Mic, indicating cell-type-specific responses in AD [19]. Moreover, male and female individuals showed differential transcriptional responses to AD pathology, especially in neurons and oligodendrocyte cells [19]. Thus, this single-cell view of transcriptional alterations associated with AD pathologies highlights the complexity of AD mechanisms from the perspective of cellular heterogeneity. Also, an understanding of neuronal vulnerability may expedite the study of circuit mechanisms in AD and provide molecular targets for circuit-specific interventions.

Glutamatergic Circuits

Dysfunction of Glutamatergic Neurons in the Hippocampus During AD

In AD, Aβ can induce neuronal hyperactivation and hypoactivation. Aβ-dependent neuronal hyperactivation may contribute to preclinical hippocampal hyperexcitability [83–85]. Mechanistically, Aβ inhibits the glutamate uptake capacity of astrocytes to produce excess glutamate accumulation, leading to the neurotoxic effects in AD [86, 87]. On the other hand, oligomeric Aβ acts on α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs) [88] and extrasynaptic N-methyl-D-aspartate receptors (NMDARs) to elicit excitotoxic effects [89, 90]. In turn, neuronal hyperactivity promotes the activity-dependent production of Aβ, eventually forming a positive feedback and leading to synapse loss, arbor shrinkage, and hypoexcitability in the late stages of AD [91–93]. In postmortem AD brains, Aβ is co-localized in glutamatergic boutons. Aβ accumulation causes loss of dendritic spines and severe neuropil damage [94–97]. Functionally, excess Aβ gains toxicity by inhibiting synaptic transmission [94, 98, 99], and impairing synaptic plasticity [94, 96, 100–103].

Tau pathology is another causal factor leading to the impairment of glutamatergic synapses. In our previous study, we found that overexpression of wild-type human tau (htau) in hippocampal CA3 decreases spontaneous excitatory postsynaptic currents with unchanged spontaneous inhibitory postsynaptic currents. These results indicate that glutamatergic synapses in hippocampal CA3 are more vulnerable to tau pathology [104]. Mechanistically, calcineurin-mediated inactivation of nuclear CaMKIV/CREB signaling is responsible for the glutamatergic impairments and memory deficits induced by htau accumulation[104]. When tau accumulates in the hippocampus, it inhibits NMDAR expression by upregulating STAT1, a transcription factor [105]. It also inhibits CREB/GluN1 phosphorylation by suppressing PKA [106]. In addition to accumulation, tau is mis-localized from the axon to the dendrites during AD progression. Then, the mis-localized tau interacts with fyn and triggers fyn to move to the dendritic spines to phosphorylate the GluN2B subunit of NMDAR, thereby enhancing excitotoxicity [107]. In AD, tau undergoes aberrant post-translational modifications, such as hyperphosphorylation, sumoylation, acetylation, glycosylation, andproteolytic cleavages, which promote tau accumulation and hippocampal propagation, causing memory deficits [108–111]. In cultured hippocampal neurons, expression of mutated tau, which mimics the phosphorylation of 14 residues, significantly reduces the number of AMPARs at synapses [112]. When expressing htau with mutations to mimic the acetylation of K274 and K281, mice have deficits in spatial and pattern separation memory with impairment in AMPAR trafficking during LTP [113]. Blocking the caspase-2 cleavage of human P301L tau by mutating the Asp314 residue on tau restores AMPAR-mediated synaptic transmission in neurons and memory deficits in mice [114]. Although P301L tau has been found to reduce the excitability of hippocampal CA1 neurons by shifting the axon initial segment (AIS) [115], Sohn et al. found that the V337M tau mutation in human neurons leads to hyperexcitability of neuronal networks by impacting the AIS cytoskeleton [116].

Recently, an in vivo neural dynamic recording system has been developed to detect cell-specific neuronal activity while monitoring the behaviors of freely-moving subjects. For example, calcium-based fiber photometry is one such system that measures calcium fluorescence signals and directly reflects neuronal spiking activity in vivo. In APP/PS1 mice, a genetically encoded calcium indicator, GCaMP6 [117], was expressed in basolateral amygdala (BLA) neurons. During the elevated plus maze test, disorganized firing patterns of BLA neurons were clearly detectable in response to BLA Aβ pathology and correlated with the anxious state of APP/PS1 mice [34]. Thus, in addition to hyperactivation and hypoactivation, AD pathologies produce abnormal neuronal firing patterns, which may result in dysfunction of information encoding and contribute to AD phenotypes.

It is worth noting that the fiber photometry system can only record population calcium. A fluorescence microscope (miniscope) system has recently been applied in neuroscience research to visualize neural activity from deep brain regions in freely-behaving animals [118]. The advantage of this calcium imaging is that the recording of neural activity from a specific population of neurons can occur longitudinally at the single-cell level of resolution. It is expected that the combination of a fluorescence microscope system and behavioral tests will provide more detailed information on neuronal dysfunction and cognitive impairment in AD.

Targeting Glutamatergic Neurons to Rescue AD-Like Memory Disorder

Whether and how targeting abnormal glutamatergic neurons could restore memory deficits remain unclear. Previous studies have suggested that episodic memory deficits in patients with AD may be due to information encoding failure [119]. However, Roy et al. employed optogenetics combined with behavioral testing and found significant disruption of memory retrieval in the early stages of AD [120]. They used immediate-early gene-dependent tools [121] to label memory engram cells (i.e., neurons holding traces of a specific memory) in APP/PS1 mice (an AD mouse model) and reported a close correlation between the progressive reduction in spine density of the hippocampal dentate gyrus (DG) engram cells and age-dependent amnesia [120]. Direct photoactivation of hippocampal DG memory engram cells with ChR2 was sufficient to induce memory recall in AD mice, indicating a deficit of memory retrieval during early AD-related memory loss [120]. Further, optogenetic induction of LTP at perforant path synapses of hippocampal DG engram cells significantly restored both spine density and memory in AD mice [120]. This study highlights the contribution of memory retrieval failure during early AD-related memory loss. Moreover, it identifies a causal role of entorhinal cortex-DG glutamatergic circuit dysfunction in memory retrieval disruption in AD.

Glutamatergic neurons in the hippocampus also receive innervation from the amygdala and their connections undergo robust degeneration during AD. Under physiological conditions, researchers combined retrograde (RV [122]) and anterograde (HSV [34]) tracing systems to delineate a novel connection between the posterior BLA (pBLA) and ventral hippocampal CA1 (vCA1), that is, the pBLA-vCA1 circuit. Using optogenetics and electrophysiological recordings, they demonstrated that vCA1 Calbindin 1 (vCA1^Calb1) neurons are downstream of pBLA glutamatergic neurons [34]. However, in AD mice, the pBLA-vCA1^Calb1 circuit is significantly inhibited in response to Aβ accumulation and results in AD-like anxiety. Photostimulation of the pBLA-vCA1^Calb1 circuit efficiently attenuates AD-like anxiety. Simultaneously, targeting the pBLA-vCA1^Calb1 circuit by optogenetics during memory retrieval significantly restores memory disorders in AD mice [34]. This novel circuit links emotions to memory. It may also provide a promising intervention target for AD patients with anxiety and cognitive impairment.

Previous work has revealed that elevated neuronal activity accelerates the progression of pathological Aβ and tau [123] and in turn continuously destroys circuit functions, forming a vicious cycle. In EC-Tau/hAPP mice, hM4Di was virally delivered and expressed in the EC, then, CNO was intraperitoneally injected. Using these chemogenetic tools, neuronal activity in the entorhinal cortex was significantly attenuated, and its beneficial effects on reducing Aβ accumulation and tau spreading into the hippocampus along the cortex-hippocampal (EC-HP) circuit were detected [77]. Thus, inhibition of hyperactivated circuits in response to Aβ and tau may be a novel strategy to prevent the progression of AD along the neural networks and break the vicious cycle between AD pathologies and memory-associated circuit dysfunction.

GABAergic Circuits

Dysfunction of GABAergic Neurons in the Hippocampus During AD

The GABA is the main inhibitory neurotransmitter in the brain. Early studies in postmortem human brains and animal models revealed that GABAergic neurons are less vulnerable than glutamatergic neurons to AD pathology. However, this idea has been challenged. There is accumulating evidence that GABAergic neurotransmission undergoes enormous changes in AD, causing excitatory/inhibitory (E/I) imbalance during AD progression.

In brain sections from patients with AD and APP/PS1 mice, perisomatic GABAergic terminals are significantly decreased, especially on the cortical neurons adjacent to amyloid plaques, suggesting an association between GABAergic neuronal impairments and Aβ [124, 125]. In addition to loss, GABAergic presynaptic terminals are elevated at an early stage in tgCRND8 and APP/PS1 mice [126, 127], indicating complicated alterations of GABAergic function in response to Aβ pathologies. During AD progression, GABAergic neuronal subpopulations are also lost. Both somatostatin ‐expressing neurons (conferring distal dendritic inhibition of pyramidal cells) and parvalbumin (PV)‐expressing neurons (providing perisomatic inhibition of pyramidal cells) are notably decreased in the brains of patients with AD and transgenic mice with Aβ pathologies [128–130].

There is a reciprocity between tau hyperphosphorylation and GABAergic synaptic dysfunction. On one hand, GABAergic signaling regulates tau hyperphosphorylation. Several GABA_A receptor modulators have been found to increase the interaction between tau and Peptidyl-prolyl cis-trans isomerase 1 (Pin 1), by which more protein phosphatase 2A (PP2A) is recruited to the cell surface to dephosphorylate GABA_A receptor β3 subunit and simultaneously reduces the availability of PP2A for tau dephosphorylation [131]. In line with this mechanism, the activation of GABA_A receptors robustly induces tau hyperphosphorylation at the AT8 epitope in cultured cortical neurons [131]. On the other hand, tau pathologies modify GABA release. In P301L mice, GABAergic interneurons are hyperactivated, resulting in higher GABA levels in the brain [132]. Thus, there may be a vicious cycle by which activation of GABA_A receptors promote tau hyperphosphorylation by reducing the association of PP2A with tau, then, the hyperphosphorylated tau enhances GABAergic neurotransmission in turn during AD progression.

Dysfunction of GABAergic transmission produces E/I imbalance in local circuits. In 14-month-old APdE9 mice, hyperexcitability was detected in the hippocampal DG and this was associated with the silencing of local inhibitory neurons [133]. Recently, we reported a prominent accumulation of hyperphosphorylated tau in a DG subset in AD patients and mice, including mature excitatory neurons, immature granular neurons, and GABAergic interneurons. However, only interneuron-specific overexpression of full-length wild-type htau to mimic AD-like tau accumulation in the mouse DG induced adult hippocampal neurogenesis deficits and increased neural stem cell-derived astrogliosis. Using calcium-based fiber photometry, hyperactivated neighboring excitatory neurons were detected in vivo after interneuron-specific overexpression of htau. Chemogenetic inhibition of excitatory neurons or pharmacologically strengthening GABAergic efficiently rescue the human tau-induced deficits in adult hippocampal neurogenesis with improved contextual cognition [134]. This work demonstrated a causal role of tau accumulation in GABAergic impairments and linked local circuit disinhibition with disruption of hippocampal neurogenesis in AD.

In addition to local microcircuits, GABAergic neurons in CA1 are also innervated by pyramidal neurons in the entorhinal cortex layer II (ECIIPN) [135]. Using a retrograde monosynaptic tracing system, researchers have identified direct connections between ECIIPN and parvalbumin-positive CA1 (CA1PV, one type of GABAergic neuron) under physiological conditions. Their functional connection (ECIIPN → CA1PV) was significantly decreased according to optogenetic and electrophysiological examinations. Further, in vivo electrophysiological recording revealed a disruption of E/I balance and dispersed place-associated firings in CA1 during ECIIPN → CA1PV degeneration in AD mice. No change in ECIIPN → CA1PN (pyramidal neurons in CA1) was observed. These data highlight the vulnerability of the ECIIPN → CA1PV circuit and its role in encoding the impairment of place information in AD.

In addition, there are some long-range inhibitory circuits from other brain regions, such as the amygdala [136] and medial septum [137], to the hippocampus. However, the details of these connections, such as the neuron type, the innervation pattern, and their transcriptome characteristics, remain elusive. More importantly, whether and how their structure and function change during AD progression also deserve further investigation.

Targeting GABAergic Neurons to Rescue AD-Like Memory Disorders

Dysfunctions in GABAergic transmission can lead to theta and gamma oscillation impairments which are important for synaptic plasticity and spatial memory. Theta oscillations (6–10 Hz) have been linked to exploratory behavior-related activities, reflecting internally-generated dynamics [138] and goal-related behaviors [139]. In rTg4510 mice (a model of tauopathy), theta and place cell (i.e. cells firing at a specific location in an environment) sequences are both significantly impaired. Principal cell firing is independent of the environmental context,, indicating the inability of mice to form new spatial memories. A reduced firing rate of the recorded interneurons may contribute to this rigid firing sequence [140]. GABAergic cell loss can decrease theta power, which is associated with cognitive dysfunction in a rat amyloid model [141]. Impaired gamma oscillation integrity has also been reported in the AD model [142, 143] and patients [144], contributing to the inefficient execution of spatial working memory [145].

However, there are distinct subtypes of interneurons in the hippocampus. Whether and how targeting different hippocampal interneurons rescues the abnormal oscillations in AD and restores memory loss remains unclear. Somatostatin-positive (SST) interneurons preferentially modulate theta oscillations [146], while parvalbumin-positive (PV) interneurons modulate gamma oscillations [147, 148]. A reduction in behaviorally-driven gamma oscillations is detectable before the onset of plaque formation or cognitive decline in a mouse model of AD. Using a Cre-dependent strategy, optoactivation of PV interneurons at gamma (40 Hz) but not at other frequencies significantly reduces Aβ levels [147]. In the J20 AD mouse model, optostimulation of PV neurons in the medial septum specifically at 40 Hz during memory retrieval restores the hippocampal slow gamma oscillation (30-60 Hz) [149]. In Aβ-injected SST-Cre or PV-Cre mice, optogenetic activation of ChR2-expressing SST and PV interneurons increases theta power and gamma oscillations, respectively, with resynchronized CA1 pyramidal cell (PC) spikes. Photoactivation of SST and PV interneurons resynchronizes SST and PV interneuron spike phases relative to theta and gamma oscillations [150], while simultaneously selectively enhancing spontaneous inhibitory postsynaptic currents onto CA1 PCs at theta and gamma frequencies, respectively [150]. Given the degeneration of the ECIIPN–CA1PV pathway in AD, optogenetic activation of ECIIPN–CA1PV synapses with a theta burst stimulation paradigm once per day for 35 consecutive days effectively enhanced theta oscillations in the hippocampal CA1 of AD mice and interrupted the progression of ECIIPN–CA1PV synaptic decays [135]. Targeting the ECIIPN–CA1PV pathway also improved the representation and precise spatial firing of CA1 place cells in AD mice, simultaneously rescuing their spatial memory deficits [135]. Together, targeting distinct interneurons may be a potential therapeutic strategy for restoring disorganized hippocampal networks and synaptic plasticity impairments in AD.

Others

The hippocampus also receives cholinergic innervation [151]. Choline acetyltransferase (ChAT) is responsible for acetylcholine synthesis and is viewed as a reliable marker of cholinergic integrity [151]. In AD brains, cortical ChAT activity is significantly decreased, and this is closely correlated with the severity of dementia [152]. In the AD hippocampus, ChAT immunopositivity is also robustly decreased [153]. Thus, degeneration of cholinergic neurons in the basal forebrain may lead to a reduction in cholinergic input to the AD hippocampus. α-Secretase-cleaved APP has been found to accumulate in cholinergic dystrophic neurites during AD progression [154]. Overexpression of Aβ decreases cholinergic spontaneous and miniature excitatory postsynaptic currents [155]. When anti-murine-p75-SAP (conjugation of saporin to a rat monoclonal antibody against the mouse p75 nerve growth factor receptor) was intracerebroventricularly injected, ChAT activity in the hippocampus and neocortex was selectively inhibited in a dose-dependent manner. These anti-murine-p75-SAP-lesioned mice showed significant spatial memory deficits in the Morris water maze test therapies remain critical in the management of patients with AD [156], the precise structural connections between cholinergic neurons in the basal forebrain and hippocampal neurons, as well as their alterations during AD progression, remain unclear.

In AD, rates of cognitive deficits are also correlated with impairment of the serotonergic system [157, 158]. Serotonergic cells are lost in the brainstem of patients with AD [159]. Selective serotonin reuptake inhibitors improve cognitive function and reduce AD-associated behavioral disorders [159, 160]. Anatomically, serotonergic neurons consist mainly of dorsal and median raphe nuclei, which send dense serotonergic projections to the hippocampus [160, 161]. It has been found that serotonin inhibits the ventral hippocampus, and its inhibition is responsible for sustained goal-directed behavior [162]. Photostimulation of 5-HT neurons expressing ChR2 in the median raphe nucleus significantly increases 5-HT concentration in the dorsal hippocampus (dHC) and produce anxiety-like behavior [163]. Further, optoactivation of 5-HT terminals in the dHC also promote anxiety [163]. Within the hippocampus, optoactivation of serotonergic terminals in CA1 enhances excitatory transmission at CA3-to-CA1 synapses and promotes spatial memory in ChR2-expressing mice [164]. Optogenetic inhibition of CA1 5-HT terminals via Arch inhibits spatial memory [164]. However, how changes occur in the hippocampus-associated serotonergic circuits in AD remain elusive.

Perspective

The study of neural circuitry is of great significance for understanding the mechanisms linking brain pathologies and cognitive symptoms in AD. Here, we have summarized the major experimental approaches for neural circuit tracing and their functional manipulations. Furthermore, the introduction of aberrant neural circuits identified in AD mouse models or patients may provide novel therapeutic strategies for AD treatment in the clinic (Fig. 3). However, some key questions need to be addressed.

Fig. 3.

Hippocampal circuits involved in AD and strategies for targeted manipulations. A Topographical dissemination of Aβ and tau pathologies in AD. Tau aggregates develop in the locus coeruleus (LC), and last in broad areas of the neocortex (NC). In contrast to tau pathology, amyloid-β deposits in AD are first observed in the NC and then in basal ganglia structures and the brainstem. B During AD progression, projections from other brain regions to the hippocampus undergo distinct alterations. Targeting those inputs may rescue spatial memory deficits in AD. C Schematic showing gradual changes of different synaptic inputs onto a CA1 pyramidal neuron across different stages of AD. Entorhinal cortex (EC) and CA3 are two major excitatory inputs, while PV and SST are two main inhibitory inputs to the CA1 pyramidal neuron. Compared with healthy brain, CA1 pyramidal neurons may become hyperactive due to a combination of decreased inhibition as well as an increase in excitation (↑E/I) during the early stages of AD. Then, during the later stages, aggressive Aβ and tau pathologies may affect synapses globally, resulting in hypoactivity (↓E/I). Targeting E/I imbalance may rescue spatial memory deficits in AD. D Stimulating engram cells in the hippocampus may improve memory retrieval in AD.

The application of viral approaches greatly facilitates research on AD neural circuit dissection. In addition to applications, we should pay more attention to their advantages and shortcomings, as noted earlier in this review. To date, there is no perfect viral vector that suits all needs. H129-ΔTK in anterograde trans-synaptic tracing cannot be used to trace local outputs because non-TK-expressing cells can also be infected at the injection site. Due to its virulence, HSV is not suitable for long-term tracing. Retrograde viral tracers, RV, also require further optimization to reduce their toxicity and meet the needs for electrophysiological recording. Furthermore, the infection and tracing efficiency of viral tracers may vary in different brain regions and cell types. Thus, cross-validating tracing results among multiple tracers is highly recommended for the dissection of AD-associated neural circuits.

Optogenetics and chemogenetics have enormous advantages in the functional studies of neural circuits. Combined with electrophysiology and tracers, we can identify functional connections between distinct neurons and evaluate their functional alterations in AD models. Together with behavioral tests, we can find specific neural circuits involved in different stages of the memory process and identify which circuit is responsible for AD-like cognitive dysfunction in various AD animal models. Optogenetics and chemogenetics have not only helped us understand the pathogenesis of AD but have also opened the door to using these tools for providing treatment. However, there are at least four major obstacles to the application of optogenetics and chemogenetics in patients with AD: (1) autoimmune reactions in patients because of the direct introduction of foreign objects into the brain, (2) the methodology of opsin and DREADD expression in specifically targeted neurons without invasion, (3) the evaluation of opsin and DREADD expression levels in specifically targeted neurons, and (4) tissue damage due to overheating from light stimulation.

Due to space constraints in this review, we introduced only some of the recent studies on hippocampal glutamatergic and GABAergic neural circuits in AD. Other glutamatergic and GABAergic neural circuits outside the hippocampal regions and neural circuits of other neuron types, such as cholinergic and serotonergic neurons, also deserve further investigation.